Hybrid multilayer perceptron-firefly optimizer algorithm for modelling photosynthetic active solar radiation for biofuel energy exploration

Modeling photosynthetic active radiation (PAR) is paramount for decisions in environmental and energy applications such as the growth of plants, agriculture, assessing oceanic conditions for coral bleaching, solar energy, and understanding of climate change/global warming and food security. This chapter develops an optimized hybrid multilayer perceptron-firefly algorithm (MLP-FFA) using advanced artificial intelligence models for PAR forecasting in regional Queensland, Toowoomba using historical lagged PAR as the model input features. The prediction capacity of MLP-FFA was compared to random forest, MLP, and multiple linear regression. Parameter tuning was performed by investigating values of (1) number of input variables; (2) train, validation and test data splits; (3) number of hidden layers; (4) hidden transfer function; and (5) training/learning algorithms, which are primary components of the MLP-FFA and MLP model. In order to select the statistically significant lagged PAR as features to be used as a model input, the partial autocorrelation function was used where a total of nine significantly lagged PAR values (i.e., t−1, t−2, …, t−9) were selected. The results show that MLP-FFA outperformed the other three models in predictive ability, elucidating the effectiveness of such a hybrid-optimized model for future renewable biofuel modeling, as well as the use of satellite data for such projects.